High Speed&High Precision Real-time Electronic Image Stabilization System Based On Embedded Platform

Electronic image stabilization system based on embedded platform combines thefeatures of high precision and portable and has been widely used in military, air surveyand photography system. This paper researches the high speed&high precision imagestabilization system based on the embedded platform according to the requirement ofthe spaceborne astro observation.Firstly，in order to evaluate different kinds of electronic stabilization algorithms’stabilization quality and the suitability of the embedded platform, this paper uses fiveevaluation indexes as stabilization precision, estimated offset range, anti-interferenceperformance, hardware acceleration and real-time performance. Considering these fiveevaluation indexes, the paper designs a evaluation method based on expectationcoefficient of electronic image stabilization algorithm used on embedded platform.Using the method, the gray projection algorithm and phase correlation are chosen asthe improved algorithms.Secondly，aiming at the shortcoming of each algorithm, two improved algorithmsare designed--the gray projection algorithm based on frequency correlation and thephase correlation based on the unite of space and frequency domain. Through thesimulations and tests, the two improved algorithms have a better performance than thetraditional ones.Traditional gray projection algorithm has a low calculation, but its precision is onlyone pixel and the estimation range is limited. Taking advantages of the FFTacceleration of embedded platform, the gray projection algorithm based on frequencycorrelation solves the problems of precision and estimation range without a largenumber of calculation.Traditional phase correlation algorithm has high precision based on complexcalculation. Using the way from careless to careful, the phase correlation based on theunite of space and frequency domain uses gray projection to get a pixel estimation anduses up-sampling by matrix multiplication to get a sub-pixel precision. The calculationof matrix multiplication is much less than the traditional zero-padding, while the twomethods have the same effect. So the instantaneity of the phase correlation algorithmbased on the unite of space domain and frequency domain is better than the traditionalone.Using the evaluation method based on expectation coefficient of electronic imagestabilization algorithm, the gray projection algorithm based on frequency correlation isselected as the core electronic image stabilization algorithm of the high speed&high precision image stabilization system based on the embedded platform.According to the requirement, the high speed&high precision real-time electronicimage stabilization system based on the embedded platform is finally designed. Thehardware architecture of FPGA+DSP is suitable to the designed system and the mainfunction modules are realized. Through the test data, the system designed can satisfythe actual demands.